<b>Charging port point-cloud recognition and positioning method based on density clustering and geometric features</b>

随着新能源电动汽车的普及和市场规模的不断扩大，电动汽车的自动充电问题逐渐成为研究热点。目前，在充电机器人领域有丰富的研究成果，其中一些已经能够实现自主充电。但是，在与充电口自动对接时，核心充电口的识别和定位存在准确率低、精度不足、计算复杂等问题，说明目前的充电口识别和定位方法存在一定的局限性。为了提高充电口识别和定位的效率和准确性，本文提出了一种新方法：基于基于密度的噪声应用空间聚类（DBSCAN）分割算法，并集成多条件约束、几何特征和双边界框：（1）利用RGB-D特征重建点云数据，并将DBSCAN算法与多条件约束相结合，将充电口区域分割成内部插座和外部半圆形点云;（2）对于分割后的点云图像，整合几何特征和双边界框，分别提取充电口的位置和方向，并进行交叉引用验证。通过在充电口不同方向平移、倾斜角度、增加物体干扰、不同光照度（0.2lux、60lux、273lux）的复杂环境下进行充电口识别和定位、精密测试、比对和对接实验，验证了该方法的鲁棒性和可行性。实验数据表明，最大位置误差为 0.061mm，最大角度误差为 0.61°，满足对接精度要求。而且，与其他研究方案相比，性能得到了一定程度的提升，为电动汽车充电的自动化提供了技术参考。With the popularization and expanding market scale of new energy electric vehicles, automatic charging for electric vehicles has gradually become a research hotspot. Currently, there is a wealth of research results in the field of charging robots, and some of them have realized autonomous charging. However, during automatic docking with the charging port, the recognition and positioning of the core charging port face issues such as low accuracy, insufficient precision, and high computational complexity, indicating that current charging port recognition and positioning methods have certain limitations. To improve the efficiency and accuracy of charging port recognition and positioning, this paper proposes a novel method: a Density-Based Spatial Clustering of Applications with Noise (DBSCAN) based segmentation algorithm integrated with multi-condition constraints, geometric features and dual bounding boxes: (1) Use RGB-D features to reconstruct point cloud data, combine the DBSCAN algorithm with multi-condition constraints, and segment the charging port area into internal socket and external semicircular point clouds; (2) For the segmented point cloud images, integrate geometric features and dual bounding boxes to extract the position and orientation of the charging port respectively, and conduct cross-reference verification. The robustness and feasibility of the proposed method are verified through charging port recognition, positioning, precision testing, comparison and docking experiments conducted in complex environments with translation in different directions, tilt angles, added object disturbances, and different illuminance levels (0.2lux, 60lux, 273lux). Experimental data show that the maximum position error is 0.061mm and the maximum angle error is 0.61°, which meets the docking accuracy requirements. Moreover, compared with other existing research schemes, the performance is improved to a certain extent, providing a technical reference for the automation of electric vehicle charging.